- UClaude Mythos 5 ubuyela emsebenzini emva kokumiswa okwethutyana kweeveki ezimbini ngenxa yokuhlolwa kweenkonzo zokhuseleko.
- Uyilo lwemodeli lubonisa inqanaba elikhulu kubuchule bokubhala iikhowudi, lufikelela kwi-80.3% kwi-SWE-bench Pro.
- Iimpawu ezintsha zememori ezihlala zihleli zivumela i-AI ukuba isebenze njenge-technical collaborator yexesha elide endaweni ye-chatbot eqhelekileyo.
- Iiprotokholi zokhuseleko ngoku ziquka indlela ephucukileyo yokubuyela umva kwiClaude Opus 4.8 kwimibuzo ebucayi.
Imeko yobukrelekrele bokwenziwa obuphezulu ibone amaza amakhulu kutshanje ngokubuya ngequbuliso kweenkqubo eziphambili ze-Anthropic. Emva kwexesha leeveki ezimbini zokumiswa ngokupheleleyo ngenxa yezizathu zokhuseleko lwesizwe, imodeli yeClaude Mythos 5 iphinde yasebenza kwiqela elithile lemibutho eyaziwa ngokuba “ngamaqabane athembekileyo.” Eli nyathelo libonisa ukuba nangona amandla ezi modeli enkulu, ukhuseleko olujikeleze ukusasazwa kwazo luhlala luqinile kunangaphambili kuluntu ngokubanzi.
Olu hlaziyo lungaphezulu kakhulu kunohlaziyo olulula lwenguqulelo; lubonisa ukujikeleza kwinto iingcali eziyibiza ngokuba yi-persistent technical collaborator. Ngokungafaniyo neenguqulelo zangaphambili ezazisoloko zilahlekelwa ngumsonto kwiincoko ezinde, uyilo lweMythos luyilelwe ukugcina ingqwalasela ixesha elide kwaye ulungelelanise imisebenzi enamanyathelo amaninzi ngaphandle kokuzibhokoxa. Kutshintsha kancinci umdlalo kwabo bafuna i-AI ukuze bahlale bekwimeko entle ngexesha leeprojekthi ezinzima, ezithatha iiveki ezininzi.
Ukusebenza okuthelekisayo kunye nobuNgcali bokuBhala iikhowudi
Xa ujonga amanani aqinileyo, inkqubela phambili iyamangalisa, ingakumbi kwicandelo lobunjineli besoftware. Kwiimvavanyo zamva nje ze-SWE-bench Pro, ezilinganisa iintloko zokubhala iikhowudi zehlabathi lokwenyani, UClaude Mythos 5 ufumene amanqaku angama-80.3%, ishiya eyandulelayo, uClaude Opus 4.8, kwisibuko sangasemva kwi-69.2%. Le ntambo enamanani amabini ibonakala kakhulu xa i-AI kufuneka idibanise iifayile ezininzi kwaye ihambe kwiikhowudi ezixineneyo nezidibeneyo.
Imiboniso yokuqala ibonise ukuba inkqubo yakha zonke ii-apps zewebhu kwifayile enye, eneetshathi ezitshintshatshintshayo kunye nemidlalo yeqonga esebenzayo enefiziksi kunye ne-logic yekhamera yeselula. Ayibhali nje ikhowudi; igqwesile kumsebenzi onzima wokulungisa iimpazamo ngokuchonga unobangela weempazamo kunye nokucebisa isisombululo esingabizi kakhulu. Kubaphuhlisi, oku kuthetha ukuba i-AI ekugqibeleni iyakwazi ukwenza ezinye zezinto ezinzima endaweni yokubonelela ngeengcebiso ezisisiseko.
Ukunyuka kwee-Arhente ze-AI eziqhubekayo
Enye yezona zinto zintle ngolu hlobo lutsha lokuphindaphinda kukukwazi kwalo ukusebenza njengearhente ezimeleyo. Endaweni yokulinda impendulo kuyo yonke indawo, Iintsomi zinokuthatha injongo ebanzi, yahlulahlule ibe ngamanyathelo afanelekileyo, kwaye ubonise isicwangciso esicwangcisiweyo ngaphambi kokuba sisetyenziswe. Imalunga nokusuka kwimibuzo ezimeleyo ukuya kwimisebenzi edibeneyo apho inkqubo ihlola kwaye igcina ulwalathiso lweprojekthi ngokwayo.
Eli nqanaba lokuzingisa lithetha ukuba imodeli inokukhumbula izigqibo zobugcisa ezenziwe kwiintsuku ezidlulileyo kwaye iphawule izicelo ezintsha ezinokuphikisana nazo. ukugcina umxholo kwiiseshoni ezinde, kuyayiphepha imigibe eqhelekileyo yokuba kufuneka wakhe kwakhona yonke ingqiqo ukususela ekuqaleni ngalo lonke ixesha uqala incoko entsha. Kufana nokusebenza nomntu osebenza naye onika ingqalelo kwimbali yomsebenzi.
Iileya zoKhuseleko kunye neeNkqubo zokuBuya
Kakade ke, ngamandla amakhulu kuza ne-red tape eninzi. I-Anthropic idibanise uthotho lwe-classifiers eziskena yonke into ngaphambi kokuba ifikelele kwi-Mythos core, ingakumbi ikhangela ii-red flags kwi-biology, i-chemistry, kunye nophuhliso oluphambili lwe-AI. Ukuba umbuzo uthathwa njengonobuzaza kakhulu okanye ufuna indlela eyahlukileyo, inkqubo isebenzisa indlela yokubuyela umva ukutshintshela umsebenzi kwimodeli enomda kancinci efana neClaude Opus 4.8.
Kubekho incoko phakathi kwabaphandi malunga nezi zihluzo zangaphakathi ngamanye amaxesha ezidala iziphumo ezingezizo ezilungileyo, nto leyo enokuba yinto ecaphukisayo xa uzama ukwenza umsebenzi. Ukulungisa oku, kukho iphulo lokuba ukucaca ngakumbi malunga nendlela ezi zihluzo ezingenelela ngayo kwinkqubo yokuyila. Ukwenza olu hlolo lokhuseleko lubonakale kunceda ekwakheni ukuthembana, ukuqinisekisa ukuba abasebenzisi bayazi kakuhle isizathu sokuba impendulo ethile iguqulwe okanye itshintshelwe kwenye indlela.
Uphuhliso lwezi nkqubo lubonisa ikamva apho amandla obuchwephesha kunye namanyathelo okhuseleko ngokusisiseko angamacala amabini engqekembe enye. Njengoko ishishini lisiya kwiimodeli ezinokuceba, ziphumeze, kwaye zikhumbule ingqiqo yazo, ukugxila kunxibelelwano olucacileyo kunye neziphumo ezithembekileyo iba yindlela ephambili yokulinganisa impumelelo. Nangona ukufikelela kusahlukile okwangoku, imilinganiselo ebekwe lolu lwakhiwo lwamva nje ibonisa indlela ecacileyo eya kwi-AI esebenza njengeqabane lokwenyani kwiimeko ezinzima zokusombulula iingxaki.
